S Simulation of data networks / fall-06. Part 2: CNCL S Simulation of data networks / CNCL.

Size: px

Start display at page:

Download "S Simulation of data networks / fall-06. Part 2: CNCL S Simulation of data networks / CNCL."

Isabel Robbins
5 years ago
Views:

1 S Simulation of data networks / fall-06 Part 2: CNCL S Simulation of data networks / CNCL CNCL: Contents CNCL C++ library for supporting event driven simulations Overview Main classes needed in simulations Learning CNCL by examples CNCL project work instructions

2 S Simulation of data networks / CNCL CNCL introduction Implemented by Communication Networks, Aachen University of Technology freeware Version 21 package can be downloaded from the course web page compiles with g++ version 295xx or less on most Unix-type platforms easy installation ( make NewWorld ) compilation on more recent compilers requires changes in Makefile(s) C++ class library collection of classes for supporting event driven simulation light weight simulation software provides functionality for example for event handling/scheduling, random number generation, statistics collection, basic statistical analysis of results Usage userwriteshisowncode(inc/c++) compiles (make) the code and links together own code and the class library S Simulation of data networks / CNCL Modeling with CNCL (1) Basic philosophy simulation model consists of processes and events processes send events to each other ev_a Process 2 ev_a Process 1 ev_a ev_c ev_c ev_a Process 4 ev_b Process 3 ev_d

3 S Simulation of data networks / CNCL Modeling with CNCL (2) Process implements a state machine (eg, a server in a queue can be in state idle/serving) receives events and depending on event s type executes an appropriate method (function) while executing the method associated with an event, it typically changes the process s state and schedules new events in practise, a process is a C++ class that has been derived (through inheritance) from the CNEventHandler class Event causes the state of a process to change events drive the simulation s execution and (usually) imply the advancement of simulation time for example, packets arriving at the queue, packet finishes service at the server, Event vs direct method call: for reasons of modularity of the program design, a process can be implemented by using several classes not all classes need to be able to handle events often the internal overhead of event scheduling can be avoided by using just a direct method call S Simulation of data networks / CNCL CNCL and C++ (1) CNCL programs implemented in C++ CNCL library provides basic functionality for pseudo random number generation, generation from given distributions, statistical analysis, basic queue elements, event management etc user can directly use these classes CNCL event handling user must implement event handling logic himself each process in the model is an event handler event handler in practise an event handler is a C++ class, that has been derived from an abstract event handler base class (CNEventHandler) event handling logic implemented in predefined functions of the derived class (the function void event_handler() is declared virtual in base class)

4 S Simulation of data networks / CNCL CNCL and C++ (2) Memory management must be handled by the user Memory space for an object can be created either statically from the stack or dynamically from the heap Stack allocation (static) { MyObject x; } an object that is declared directly is created on the stack and objects are destroyed in reverse order of creation (done automatically on exit of the block they were created in) user has no control over the creation timing and destroying eg random number generators and event schedulers can be such static objects created in the main program S Simulation of data networks / CNCL CNCL and C++ (3) Heap allocation (dynamic) { MyObject* xptr; xptr = new MyObject; delete xptr; } objects created with the new operator are placed on the heap and will persist until explicitly destroyed, or the program terminates every object created with new must be explicitly destroyed with a corresponding delete eg in simulation of a network (or just single queue) packets going through the system should be dynamic objects

5 S Simulation of data networks / CNCL CNCL properties Pros as the user implements all functionality, the user also has full control of what functionality is needed and what is not fast execution times (no unnecessary overhead) (relatively)easytolearn(simple) good support for random number generation and event driven simulation Cons no ready made functional blocks for network simulations (eg, different protocols, etc) implementation time may be substantial S Simulation of data networks / CNCL CNCL: Contents CNCL C++ library for supporting event driven simulations Overview Main classes needed in simulations Learning CNCL by examples CNCL project work instructions

6 S Simulation of data networks / CNCL Essential functionality needed in every simulation program To some extent, all simulations need the same basic building blocks pseudo random number generator random number generators from given distributions event scheduler (event exploders, ) different queues (FIFO, priority queues, ) event data structure jobs (packets) S Simulation of data networks / CNCL Class hierarchy Class CNCL provides functions for error handling CNObject is the root of the CNCL class hierarchy CNCL CNObject CNRNG CNRandom CNStatistics CNEvent CNEvent Handler CNJob

7 S Simulation of data networks / CNCL Random number generators CNRNG is an abstract base class for all CNCL random number generators the pseudo-random number generators to be used have been derived from CNRNG CNRNG supports following pseudo random number types unsigned integer float 01 double 01 Actual RNGs differ in quality of pseudo random number sequences (sequence lengths, overlapping sequences, correlation), efficiency, and memory consumption S Simulation of data networks / CNCL Derived classes (1) CNLCG Linear Congruence RNG simplest pseudo random number generator may be used when performance is more important than perfect randomness CNMLCG Multiple Linear Congruence RNG combines the results of two different CNLCGs implementation taken from the GNU-library libg++ fairly long period, and has been shown to give good intersample-independence CNACG Additive RNG high quality random number generator requires a fair amount of memory for each instance of the generator implementation taken from the GNU-library libg

8 S Simulation of data networks / CNCL Derived classes (2) CNFiboG Fibonacci RNG high quality generator with a huge period (in the CNCL implementation period = ) relatively high memory usage CNFileG Data File RNG data file random number generator class reads random numbers from a disk file good file must have a sufficient size considerable memory usage and low speed can be expected when using this class CNTausG Tausworth RNG main advantage of this generator is that it can easily be implemented as a fast hardware generator statistical tests have shown some flaws in this generator so that its use is not recommended S Simulation of data networks / CNCL Summary of random number generators Randomness Period length Performance Memory req LCG MLCG ACG +++ 1) ++ + FiboG ) depends on table size

9 S Simulation of data networks / CNCL Example CNFiboG rng1; CNRNG *rng2 = new CNFiboG(); unsigned x1; float x2; double x3; x1 = rng1as_long(); // draw a random integer 02^31-1 x2 = rng2->as_float(); // draw a random float 01 x3 = rng2->as_double(); // draw a random double 01 delete rng2; S Simulation of data networks / CNCL Random numbers with different distributions CNRandom is an abstract base class for different random number distributions common interface to access all derived RNG classes CNRandom provides a random number from the distribution CNRandom uses CNRNG CNRandom initialized with a pointer to the used pseudo random number generator

10 S Simulation of data networks / CNCL Derived classes CNBeta CNBinomial CNDeterm CNDiracTab CNDiscUniform CNErlang CNGeometric CNHyperExp CNHyperGeom CNInterTab CNLogNormal CNMDeterm CNNegExp CNNormal CNPoisson CNRandomMix CNRayleigh CNRice CNTab CNUniform CNWeibull S Simulation of data networks / CNCL Example CNRNG *rng = new CNFiboG(); double mean = 20; CNNegExp rnd(mean, rng); double x; x = rnd(); // draw a neg exp distributed random number delete rng;

11 S Simulation of data networks / CNCL Statistical evaluation CNStatistics is an abstract base class for all statistics classes defines a common interface CNStatistics allows to put a value for statistical evaluation to reset the evaluation CNStatistics provides eg mean and variance of the input sequence number of evaluated values minimum and maximum of all evaluated values Derived classes CNMoments CNMomentsTime CNConfidence CNHistogram (CNLREF, CNLREG, CNDLRE, CNBatchMeans) S Simulation of data networks / CNCL Evaluation of moments for simulation data CNMoments provides eg mean variance and relative variance 2nd and 3rd zero moment 3rd central moment relative deviation skewness CNMomentsTime moments of a time-weighted input sequence useful for computing eg statistics of the queue length process CNConfidence usual non-parametric statistics + functions for computing confidence intervals CNHistogram support for computing histograms of sample statistics

12 S Simulation of data networks / CNCL Example(s) CNMoments m; double x; double y; CNMomentsTime m; double x; double y; mput(20); mput(34); mput(51); mput(20, 10); mput(34, 20); mput(51, 30); x = mmean(); y = mvariance(); x = mmean(); y = mvariance(); cout << m; cout << m; S Simulation of data networks / CNCL Container classes Container classes work with pointers to CNObject Generic data structures CNAVLTree AVL balanced tree structure CNSLList Single Linked List of Objects CNDLList Double Linked List of Objects also iterators for lists Queue objects CNQueueFIFO FIFO Queue CNQueueLIFO LIFO Queue CNQueueRandom Random queue CNQueueSPT Shortest Processing Time queue (only for CNJobs) CNPrioQueueFIFO FIFO priority queue Other classes CNSink Queue that deletes all inserted jobs

13 S Simulation of data networks / CNCL FIFO queue CNQueueFIFO queue; CNJob* in_job = new CNJob; CNJob* out_job; queueput(in_job); out_job = queuepeek(); //job not removed from queue out_job = queueget(); //job is removed from queue delete out_job; S Simulation of data networks / CNCL Job CNJob (derived from CNObject) is a standard object for CNCL queues CNJob provides eg the following public member variables: CNSimTime in // enter system CNSimTime start // service begins CNSimTime out // leave system int priority // priority of job Useful for example when recording sojourn times

14 S Simulation of data networks / CNCL Example CNMoments m_queue; CNMoments m_total; CNJob *job = new CNJob; job->in = now(); // job arrives at the queue job->start = now(); // service begins job->out = now(); // service ends m_queueput(job->start - job->in); m_totalput(job->out - job->in); delete job; // job is no longer needed S Simulation of data networks / CNCL Event driven simulation Event Handlers: Generator, Server Events: TIMER_G, TIMER_S, JOB Event Scheduler Scheduler TIMER_G TIMER_S JOB JOB Generator Server

15 S Simulation of data networks / CNCL Event handlers Derived from class CNEventHandler state machine that receives and processes events user implements event handling method: void event_handler(const CNEvent *ev) executed routine depends on the state of the event handler and the type of the incoming event may generate new events and change state resembles a process S Simulation of data networks / CNCL Example Class Generator : public CNEventHandler { private: public: }; void Generator::event_handler(const CNEvent *ev) { switch (ev->type()) { } };

16 S Simulation of data networks / CNCL Events Class CNEvent data structure representing events in the simulation Includes type priority sending time and scheduled time sending and receiving event handlers unique identifier pointer to an arbitrary CNCL object Note! CNEvents are created with new operator, but they do not need to be explicitly deleted by user (scheduler takes care of that) S Simulation of data networks / CNCL Example CNEvent *ev; CNRandom *rnd; Server *server; // another event handler ev = new CNEvent(EV_TIMER_G); send_delay(ev, rnd( )); // send to myself as default ev = new CNEvent(EV_JOB, server, new CNJob); send_now(ev); // send without delay send_delay(new CNEvent(EV_TIMER_G), rnd( ));

17 S Simulation of data networks / CNCL Event scheduler Operation controls the simulation run receives events orders events in the increasing order of time stamp and decreasing order of priority passes events to the addressed event handlers deletes all created CNEvent objects Two variants CNEventScheduler guarantees that events are processed in FIFO order even when time and priority compare equal slow if nof managed simultaneous events grows large CNEventHeapSched more efficient than CNEventScheduler, but cannot guarantee FIFO processing if time and priority of events are equal S Simulation of data networks / CNCL Examples main( ) { CNEventScheduler scheduler; schedulerstart(new CNEvent(EV_TIMER_G, &generator, 0)); } main( ) { CNEventScheduler scheduler; schedulersend_event(new CNEvent(EV_TIMER_G, &generator, 0)); schedulerstart( ); }

18 S Simulation of data networks / CNCL Simulation time Note! Simulation time is accessible only in EventHandler classes in fact, only in the EventHandler method!!!! Within such classes, simulation time is accessed with now() method call each event handler has inherited this method from the base class CNEventHandler the method now() works properly only within the EventHandler -method

Simulation Studies of the Basic Packet Routing Problem

Simulation Studies of the Basic Packet Routing Problem Author: Elena Sirén 48314u Supervisor: Pasi Lassila February 6, 2001 1 Abstract In this paper the simulation of a basic packet routing problem using